dcc-mcp-PowerPoint

PowerPoint adapter for the DCC-MCP ecosystem — the application layer over dcc-mcp-office: deck generation from structured content, slide composition, review decks from DCC renders, batch PDF conversion and text replacement (via the shared core), previews and visual validation.

Status: M1 + M2 authoring live. The designed pipeline runs end-to-end: Deck IR (office-ir/1.0) → Open XML compile (python-pptx) → desktop COM render (PowerPoint DispatchEx, dedicated instance) → PDF + per-slide PNG previews → structural validation. On top of generation, M2 adds structured deck modification (ppt-patch/1.0), smart semantic layers and a plugin registry — see the skill packs below. Skill scripts are executable through the dcc-mcp gateway (DCC_MCP_POWERPOINT_SKILL_PATHS). The shared C# host in dcc-mcp-office remains the production COM path (M1 there).

Part of the DCC-MCP host matrix

dcc-mcp-PowerPoint — PowerPoint adapter for DCC-MCP — deck generation (Deck IR → Open XML compile → COM render), slide composition and review decks.

It is one of 47 host adapters in the DCC-MCP catalog. Every adapter speaks the same MCP protocol and builds on the same core runtime contract; each one exposes the tools its own host needs on top of that.

• All host adapters and install metadata
• Host matrix on the core README
• Showcase

This block is generated from the catalog entry in dcc-mcp-catalog.yml. Re-run the generator after changing the catalog.

Design (two-layer sidecar)

Agent → dcc-mcp-gateway (Rust, dcc-mcp-core)
          → dcc-mcp-server sidecar  (lifecycle: RFC #998)
              → office-host.exe --app=powerpoint  (C#, dcc-mcp-office)
                  → POWERPNT.EXE (COM, interactive user session)

• Gateway + sidecar lifecycle: existing dcc-mcp-core machinery.
• office-host: shared C# runtime from dcc-mcp-office (download/verify/launch via dcc-mcp-release-artifacts).
• This repo owns: PowerPoint IR semantics, powerpoint.deck.generate / powerpoint.slide.compose / powerpoint.slide.render, template registry entries, and the PowerPoint skill packs.

Skill packs

• powerpoint-deck — generate an editable deck from structured content (template → semantic layouts → IR → Open XML → COM finalize → previews → validation loop; proposal §15.3/§15.4). Generated decks are tagged with six semantic layers (background/decoration/header/content/accent/footer).
• powerpoint-edit — modify existing decks with a structured patch document (ppt-patch/1.0): text, notes, pictures, fills/lines, visibility, shape deletion, slide add/delete/move (new slides compile from Deck IR layouts) and layer assignment. All-or-nothing saves.
• powerpoint-layers — smart semantic layers over shapes (native ::layer=<name> name tags): list, show/hide, restack z-order (front/back/above/below) and recolor per layer.
• powerpoint-plugins — discover and invoke user-installed plugins (plugin.json manifest + stdin-JSON script contract, subprocess isolation, timeout caps).
• powerpoint-review — dcc.review-deck-from-renders: build review decks from Maya/Houdini/Blender/Unreal renders + shot/asset metadata (proposal §15.7).

Capability-boundary showcase

examples/capability_boundary_showcase.json is a 17-slide deck that walks every semantic layout and the M2 capability surfaces: generation (IR → Open XML → COM render → validation), smart layers (list / visibility / recolor / reorder / assign), the patch engine (7-operation ppt-patch/1.0 with all-or-nothing saves) and the plugin registry (deck-stats demo plugin in examples/plugins/). See examples/output/capability-boundary/README.md for the full runbook and artifact map.

Agent usage (via dcc-mcp-cli / gateway)

1. Start the registered adapter (the launcher binds the bundled skill packs and script runtime to the official dcc-mcp-server): dcc-mcp-powerpoint serve
2. Run a skill script directly (gateway execute_script contract — stdin JSON or CLI flags):

   python src/dcc_mcp_powerpoint/skills/powerpoint-deck/scripts/generate_deck.py \
     --input examples/dcc_mcp_framework_intro.json --out out
   

3. Validate the packs with the official linter:

   dcc-mcp-cli lint src/dcc_mcp_powerpoint/skills --warnings-as-errors --non-interactive
   

Development

pip install -e .[dev,sidecar]
pytest

License

MIT